An Experimental Approach to Detect Forest Fire Using Machine Learning Mathematical Models and IoT

Fire outbreak is a common issue which is occurring worldwide, causing significant damage to both nature and human life. Recently, vision-based fire detection systems have gained popularity over traditional sensor-based systems. However, the detection process using image processing techniques can be tedious. In the current study, we propose a technique for fire detection that utilizes optimal convolution neural networks (OPCNN) to achieve highly accurate detection of fire images in forest. The result of proposed model is compared with two other models: CNN and J48. The proposed model performs better than these models. The proposed algorithm was trained using a dataset consisting of 755 images of fire and 244 images of non-fire, for a total of 999 images. These images were obtained from Kaggle data set. We resized and reshaped 1380 of these images for use in training and 460 images for testing. The model was trained using convolution, activation functions, and max pooling operations with different batch sizes and epoch values. The resulting model achieved an accuracy of 95.11%, with 432 out of 460 images predicted correctly. The proposed approach thus provides a highly accurate and efficient method to detect forest fire accurately for a sustainable safety world. The proposed work provides a new direction towards accurate and early detection of fire not only in forest but also in case of agriculture field, rural, urban, and many more areas.